Headroom: A Context Compression Layer for AI Agents
An introduction to Headroom, a context compression layer for AI agents and LLM applications that reduces token usage before tool outputs, logs, files, and RAG chunks reach the model.
OpenClaw Practice
Introductory guides, skills, and practical notes around OpenClaw.
Related articles
Sorted by publish date, with content strongly related to this topic first.
→
html-video: Generate Real MP4s Locally with HTML, CSS, and Coding Agents
html-video is an open-source project from the Open Design team. It lets coding agents understand content, build storyboards, express frames with HTML/CSS, and render real MP4 videos locally through Chromium and ffmpeg.
html-ppt-skill: Let AI Generate Professional HTML Presentations
html-ppt-skill is an AgentSkill for generating professional HTML presentations with themes, layouts, animations, and presenter mode using pure static HTML/CSS/JS.
Agents Need More Than Models: Context Infrastructure Is Taking Shape
OpenViking, Mirage, and SkillOpt show that AI agent systems are moving beyond model calls toward context databases, virtual filesystems, and optimizable skills.
The Free Era of AI Tools Is Ending: From Copilot Token Billing to Agent Budget Governance
GitHub Copilot is moving to usage-based billing, OpenRouter has launched Guardrails, and projects like AgentBudget point to a broader shift from subscription-style AI usage to metered, governed agent workflows.
用 Codex 优化网速靠谱吗？先把它当成网络诊断助手
从 X 上流传的 Codex 优化网速案例和爱范儿实测说起：Codex 可能帮你发现 DNS、Wi-Fi、代理、后台进程和系统网络配置问题，但它不是一键突破带宽上限的加速器。
Pi: A Coding Agent Harness You Can Reshape Around Your Workflow
An introduction to earendil-works/pi: a terminal-first, extensible AI coding agent harness with a CLI, agent runtime, unified model API, TUI components, and package system.
What GitHub Trending Says About AI Agents: Skills Are Becoming Engineering Assets
A look at Understand-Anything, Superpowers, ECC, knowledge-work-plugins, stop-slop, taste-skill, and other GitHub Trending projects that point to AI agents moving from prompts to installable, reusable, auditable work methods.
AI Agents Need Observability: Why Security Logs Must Capture Intent to Action
AI agent security logs should not only record whether an API call succeeded. They need to connect user intent, agent planning, tool calls, policy decisions, human approval, and execution results.
Codex Is No Longer Just a Coding Tool. It Is Becoming a Long-Running Task Workbench
Codex is moving beyond code generation into long-running work: durable threads, goals, tool-backed verification, automations, and project rules make tasks easier to continue, inspect, and finish.
HTML Anything: Let Local AI Agents Generate Publish-Ready HTML
HTML Anything is a local-first agentic HTML editor that turns Markdown, tables, drafts, and data into publish-ready single-file HTML or PNG using the AI coding agents already logged in on your machine.
Marketing Skills for AI Agents: A Marketing Workflow Pack for AI Agents
Marketing Skills for AI Agents is a marketing skill pack for AI coding agents, giving Claude Code, OpenAI Codex, Cursor, Windsurf, and other agents more structured workflows for SEO, conversion optimization, copywriting, ads, analytics, and growth tasks.
Persistent Memory Is Not About Remembering Everything: What Should Agents Actually Keep?
Long-term memory for AI agents should not be a dumping ground for chat history. It should be a verifiable, scoped, and maintainable layer of working context that changes future behavior.
Codex Chrome Extension: When Should AI Use Your Browser Login State?
The Codex Chrome extension lets Codex work with Chrome when a task depends on signed-in browser state, real web pages, or browser-context verification.
Warp: An AI Agent Terminal with Built-in Code Review Workflow
Warp is an AI-native development environment born from the terminal, combining the command line, coding agents, Git diff review, code review, and commit workflows in one workspace.
open-slide: A Slide Framework Built for AI Agents
An introduction to open-slide, a React slide framework for AI agents where you describe the deck in natural language, agents write the React pages, and the framework handles preview, presentation, export, and iteration.
Keep Codex Fast: A Safe Maintenance Skill for Keeping Local Codex State Lightweight
An introduction to keep-codex-fast, a local maintenance skill for heavy Codex users. It emphasizes inspection, handoff notes, backups, and safe archiving before touching old sessions, logs, or workspaces.
HyperFrames: An Open-source Rendering Framework for Generating Video with HTML
An introduction to HyperFrames, HeyGen's open-source framework for defining videos with HTML, CSS, and frontend animation tools, then rendering them to MP4 through a browser and FFmpeg.
Multica：把编码 Agent 当成团队成员管理的开源平台
Multica 是一个开源 Managed Agents 平台，用看板、Issue、运行时和技能系统，把 Claude Code、Codex、OpenCode、Hermes 等编码 Agent 纳入团队协作流程。
Codex 使用最佳实践：别把它当聊天机器人，要当成一个可配置的工程队友
基于 OpenAI Codex best practices 整理：如何通过上下文、Plan、AGENTS.md、配置、测试验证、MCP、Skill、Automation 和 session 管理，更稳定地使用 Codex。
Obscura：一个为 AI Agent 和网页抓取而生的 Rust 无头浏览器
介绍 h4ckf0r0day/obscura：一个用 Rust + V8 实现、兼容 Puppeteer/Playwright 的轻量无头浏览器，面向 AI Agent 和网页抓取场景。
Browser Harness：让 AI 直接接管真实浏览器的轻量自愈执行层
介绍 browser-use/browser-harness：一个基于 CDP 的轻量浏览器执行层，支持 AI 在任务中直接连接真实浏览器，并在执行过程中补齐缺失能力。
Manifest：帮个人 AI Agent 降低模型成本的开源路由器
介绍 mnfst/manifest：一个面向个人 AI Agent 的开源模型路由器，支持按请求复杂度自动选模型、设置回退链路，并跟踪 token、成本与用量。
Superpowers：给 Claude Code、Codex、Cursor 装上一套更靠谱的开发方法论
Superpowers 不是新的 Coding Agent，而是一套面向 Claude Code、Codex、Cursor 等工具的 skills 框架和开发方法论。它想解决的，不是 AI 会不会写代码，而是 AI 能不能按工程流程把项目稳定推进下去。
fireworks-tech-graph 项目解析：把技术架构图生成这件事做成一个可复用 Skill
介绍 yizhiyanhua-ai/fireworks-tech-graph 这个 Claude Code Skill：它如何把自然语言描述转成可直接发布的 SVG + PNG 技术图，以及它适合哪些技术文档场景。
从 OpenClaw 到 Hermes Agent：为什么我觉得它更懂你
结合实际迁移体验，记录 Hermes Agent 的安装、配置、Telegram 接入，以及它为什么比 OpenClaw 更容易形成“更懂你”体验的技术原因。
mem9 平台介绍：和本地 memory 对比后，你就知道它值不值得上
先讲 mem9 与本地 memory 的核心差异，再讲场景与原理，帮助你判断是否该接入。
Hermes Agent 项目解析：一个能持续学习的开源 AI Agent 框架
深入解析 NousResearch/hermes-agent 的定位、能力边界、适用人群与可变现方向，帮助你判断是否适合落地。
OpenClaw Skills 去哪找？我常用的 6 个渠道与实用命令
整理 OpenClaw Skills 的高质量来源与筛选方法：ClawHub、官方文档、GitHub、社区实战，以及一套避免踩坑的选型流程。
AiToEarn：多平台内容增长与变现一体化工具
用模板化方式拆解开源项目 AiToEarn：基础信息、核心能力、适用人群、优缺点与可变现方向。
